JOURNAL OF WOMEN’S HEALTH Volume 00, Number 00, 2016 ªMary Ann Liebert, Inc. DOI: 10.1089/jwh.2015.5529
Abstract
The Association of Inflammation with Premenstrual Symptoms
Original Article
Ellen B. Gold, PhD, Craig Wells, BA, and Marianne O’Neill Rasor, MA
Background:About 80% of women experience premenstrual symptoms (PMSx), and about 50% of women seek medical care for them, posing a large medical care burden. However, despite women’s use of anti inflammatory agents for relief from these symptoms, and the fact that antiinflammatory agents provide relief from some PMSx, the relationship of inflammation to PMSx has not been well investigated. Methods:We, therefore, undertook the present crosssectional analyses using baseline data from the longitu dinal Study of Women’s Health Across the Nation (SWAN), a racially/ethnically diverse cohort of midlife women (n=2939), to determine if a biomarker of inflammation, highsensitivity Creactive protein (hsCRP), was associated with PMSx. We performed factor analyses with Varimax rotations to determine five groupings of eight symptoms to develop a parsimonious set of outcome variables. We conducted backward stepwise multiple logistic regression models for each grouping, eliminating nonsignificant (p>0.05) covariates. Results:Having an hsCRP level>was significantly positively associated with premenstrual mood3 mg/L symptoms (adjusted odds ratio [aOR]=1.27, 95% confidence interval [95% CI] 1.02–1.58), abdominal cramps/ back pain (aOR=1.40, 95% CI 1.09–1.80), appetite cravings/weight gain/bloating (aOR=1.41, 95% CI 1.04– 1.89), and breast pain (aOR=1.26, 95% CI 1.02–1.55). Elevated hsCRP level was not associated with pre menstrual headaches or reporting three or more PMSx. Conclusions:The significant relationships of specific groups of PMSx with elevated hsCRP levels have potential clinical implications for treatment and possibly for prevention by advising women about the factors associated with inflammation and the potential for treatment with antiinflammatory agents.
Introduction
remenstrual symptoms (PMSx) includemood, P physical, and cognitive symptoms that begin in the luteal phase of the menstrual cycle and end with, or shortly after, the 1 onset of menstruation. The frequency, type, severity, and 2 combination of symptoms that comprise PMSx vary. The most frequently reported symptoms are irritability, depres sion, fatigue, water retention, weight gain, breast tenderness, 3 headaches, abdominal cramps, and mood swings. About 4 80% of women may experience PMSx, and about 50% of 5–7 women seek medical care for them, thus posing a large medical care burden. The etiology of PMSx may be related to ovarian function, as suppression of ovarian hormone secretion markedly at 8 tenuates PMSx, although differences in ovarian steroid hormones have not been consistently observed between symptomatic and asymptomatic women. Biologic, social, demographic, and behavioral factors have been inconsis 2,9–12 tently associated with PMSx.
Highsensitivity Creactive protein (hsCRP) is an acute phase inflammatory marker that has been associated with 13 cardiovascular disease risk and is an outcome associated 14 with menopausal vasomotor symptoms. It has also been associated with some of the risk factors for PMSx, such as smoking, depressive symptoms, increasing age, and in 14 creased body mass index (BMI). While some studies have investigated the associations of inflammation with PMSx, most of these have had relatively small samples of young (e.g., 15,16 ages 18–30 years) white women, and have found sug gestive, but not always significant differences in inflammation between women reporting and women not reporting emotional or physical PMSx. Furthermore, antiinflammatory agents have been found to 17 provide relief from some PMSx. It is thus possible that inflammation is the mechanism by which these factors in crease the risk of PMSx. Therefore, establishing the role of inflammation in different types of PMSx in a large diverse sample of women would be informative in understanding the potential physiologic mechanisms involved in PMSx. We
Department of Public Health Sciences, School of Medicine, University of California Davis, Davis, California.
1
2
undertook these crosssectional analyses of PMSx among a racially/ethnically diverse cohort of midlife women to de termine if inflammation, as measured by hsCRP, was asso ciated with PMSx.
Methods Study participants This crosssectional study used data on PMSx, health, re productive, demographic, and lifestyle factors from the baseline questionnaires of the Study of Women’s Health Across the Nation (SWAN), a longitudinal, multicenter, multiracial/ethnic study of midlife women. SWAN is fol lowing a cohort of women (N=3302 at baseline) from five racial/ethnic groups, at seven clinical sites located nation 18 wide. We recruited communitybased cohorts of Cauca sians and one nonCaucasian group at each site: African Americans in Pittsburgh, Boston, Detroit, and Chicago; Hispanics (Puerto Rican, Dominican, Cuban, Central and South American) in Newark, New Jersey; Japanese in Los Angeles; and Chinese in the Oakland, California area. Participants were eligible for inclusion in the cohort if they were aged 42–52 years and pre or early perimenopausal, had not undergone a hysterectomy or bilateral oophorectomy, were not pregnant, and were not using menopausal hormone therapy or oral contraceptives at baseline. In addition, par ticipants were required to be able to speak English, Spanish, Cantonese, or Japanese, and to provide informed consent to participate and comply with the study protocol. All instru ments and the study protocol were approved by the institu tional review boards at all sites, and signed, written informed consent was obtained from all study participants. From the total baseline sample of 3302 women, 57 were excluded for missing Creactive protein (CRP) data; 129 additional women were excluded for missing data on PMSx; and an additional 2 women were excluded for missing in formation on whether the symptoms disappeared within 3 days of onset of their menstrual period.
Data collection All SWAN participants completed a selfadministered and intervieweradministered questionnaire at baseline.
Outcomes.These analyses included data from the base line visit (administered during 1996–1997) at which partici pants indicated yes or no in response to the following question for each of eight symptoms: ‘‘During the last year, have you had any of the following during at least half of your menstrual periods or in the week before them?’’ The eight symptoms included the following: abdominal cramps/pain, breast pain/tenderness, weight gain/bloating, mood changes/ suddenly sad, increase in appetite or cravings, anxious/jittery/ nervous, back/joint/muscle pain, and severe headaches. If a participant answered yes to any one of the symptoms, she was also asked the following question: ‘‘Did this/these characteristic(s) usually (more than half of the time) disap pear within 1–3 days after your period started?’’ Answering ‘‘yes’’ to this question was used as the criterion for a symp tom to be considered premenstrual in the present multivariate analyses. Those who answered ‘‘no’’ or ‘‘don’t know’’ were excluded from multivariate analyses (an additional 175 who
GOLD ET AL.
reported symptoms answered no or don’t know to whether the symptoms disappeared within 3 days of onset of their men strual periods; so, the total number excluded=363 when using this more conservative definition of PMSx, but only 188 were excluded if the more expanded criteria were used of reporting the symptom, but saying no or don’t know in re sponse to whether the symptom disappeared within 3 days of onset of their menstrual periods).
Independent variable.hsCRP assays were performed at baseline using an ultrasensitive rate immunonephelometry (hsCRP on BN100; DadeBehring, Marburg, Germany). The method is based on monitoring light scattering during ag glutination of CRP to polystyrene particles coated with monoclonal antibodies to CRP. The sensitivity of the assay (lowest detectable concentration) was 0.03 mg/dL. The in terassay coefficients of variation at CRP concentrations of 0.05 and 2.2 mg/dL were 10%–12% and 5%–7%, respec tively. Although hsCRP level is a continuous variable, a 19 cutoff for elevated hsCRP has been established for clinical use and was used to categorize hsCRP into elevated (>3 mg/L) and nonelevated (£for analyses.3 mg/L)
Covariates.Age at baseline was analyzed as a continuous variable. Annual household income was selfreported and evaluated using a threelevel categorical variable based on tertiles of total income reported<$35,000, $35,000–$75,000, and>$75,000. A binary categorical variable was used for the proportion of women with a college education. Race/ethnicity was selfidentified as Caucasian, African American, Hispanic, Chinese, or Japanese and included both USborn and foreign born women. Menopausal status at baseline was defined using a di chotomous variable: (1) premenopausal (menstrual period in the prior 3 months with no change in regularity of periods) or (2) early perimenopausal (menstrual period in the prior 3 months with change in regularity of periods) without use of hormone therapy. Parity was selfreported and analyzed as a categorical variable. Weight and height were measured using a calibrated bal ance beam scale and stadiometer, respectively. BMI (weight in 2 kilograms/[height in meters] ) was computed and analyzed as a fourlevel categorical variable: low (<18.5), normal (18.5– 24.9), overweight (25–29.9), or obese (‡30). Comorbidity consisted of reporting of 1 or more of 10 chronic health con ditions (heart disease, arthritis, high blood pressure, diabetes, high cholesterol, stroke, anemia, migraines, angina, and oste oporosis) during the past year and was treated as a categorical variable. Use of antiinflammatory medications was assessed by selfreported use in the prior month of such prescription and nonprescription medications as assessed by SWAN pharma cologists, independent of report of PMSx. Active smoking status was assessed by standard ques 20 tions. Passive smoke exposure was assessed by the vali 21 dated instrument of Coghlinet al.Never smokers with no passive smoke exposure were used as the referent group. Physical activity was measured by a composite score 22 based on the Kaiser Permanente Activity Score, a modifi 23 cation of the Baecke scale assessing three domains: sports, leisure, and household activities. Usual servings of alcoholic beverages per week were analyzed as none,£1, and>1 (one serving=beer, 5 oz. wine, or 1.5 12 oz. oz hard liquor).
INFLAMMATION AND PREMENSTRUAL SYMPTOMS
Social support was assessed by a summed scale of how often four types of needed emotional and instrumental supports were available, with responses ranging from 0=none of the time to 24 4=and analyzed by quartiles of the total scoreall of the time in the SWAN baseline cohort. A measure of the symptom sensitivity trait was measured at followup visit 01 using a summed score (degree of awareness of loud noise, hot or cold, hunger, pain, and things happening in one’s body, with re 25 sponses ranging from 1=not at all true to 5=extremely true) and analyzed dichotomously as at or above versus below 15, the median for the SWAN cohort. Depressive symptoms were assessed by the Center for Epidemiologic Studies Depression 26 (CESD) scale (score‡16 on a 20item scale of the extent to which each item was experienced in the previous week).
Data analyses This was a crosssectional analysis, using only data from the baseline visit. Descriptive statistics were computed using bivariate analyses for each symptom grouping (as described below), each independent variable, and each covariate. Ca tegorical variables were analyzed using chisquare tests or Fisher’s exact test for comparison of proportions, andttests and analysis of variance (ANOVA) were used for compari sons involving continuous variables. Unadjusted odd ratios (ORs) were computed for each symptom group by each in dependent variable. We conducted factor analyses with Varimax rotations to determine appropriate groupings of the eight symptoms so that a parsimonious set of outcome variables could be eval uated. To determine whether to retain a particular symptom in a symptom grouping, we used factor loadings of 0.40 or more. If items loaded on more than one factor, the item with the highest loading was retained. Factors were accepted with 12 an eigen value of 1.0 or greater. As in our prior work, the five resulting PMSx groupings were as follows: (1) anxiety/ jittery/nervous and mood changes, (2) abdominal cramps and back/joint/muscle pain, (3) increased appetite/craving and weight gain/bloating, (4) breast pain/tenderness, and (5) headaches. Because women often reported more than one symptom, associations of the independent variables with the total number of these five symptom groupings (>3 vs.£3) were also estimated. To assess potential confounding variables, we calculated unadjusted odds ratios (ORs) and 95% confidence intervals (95% CIs), one variable at a time. To adjust simultaneously for confounding variables, multiple logistic regression models were developed for each PMSx grouping. Covariates that were associated (atp<0.15) in unadjusted analyses were entered into backward stepwise multiple logistic regression models for each PMSx grouping with elimination of variables found not to be significant (p>0.05). The independent variable, elevated hsCRP (>3 mg/L vs.£was forced into all multiple3 mg/L), logistic regression models. AIC goodness of fit test criteria were used for multiple logistic regression models. Interactions with raceethnicity and menopause status were evaluated to deter mine if any relationships observed differed by these variables.
Results The unadjusted proportion of women who reported each PMSx, except breast pain or headaches, was significantly in creased for women who had hsCRP values>3 mg/L (Table 1).
3
In addition, mean age was significantly lower among women who reported all PMSx except for those reporting premen strual breast pain. All symptoms were reported by significantly more Hispanics and early perimenopausal women and by significantly less Chinese and Japanese than Caucasian or premenopausal women. Most symptoms (except changes in appetite/weight/bloating and breast pain) were reported by fewer women with more than a high school education, higher annual income, and lower symptom sensitivity scores com pared to those with a high school education or less, lower annual income, and higher symptom sensitivity. Most symptoms (except for breast pain or headaches) were reported by significantly more obese women, those with ac tive or passive smoke exposure, and by women with elevated depressive symptom scores (for all symptoms) than normal weight women, women without active or passive smoke ex posure, or women with lower depressive symptom scores. Parity, physical activity, hypertension, arthritis, and anemia were significantly positively and alcohol consumption was significantly negatively related to headaches. However, most of the differences were relatively small and likely significant because of the large sample size. Diabetes, cancer, high cholesterol, stroke, and thyroid disease were not significantly related to any symptoms, nor was heart disease except for a significant relationship to abdominal cramps and pain.
Unadjusted analyses In unadjusted analyses, hsCRP levels>were sig3 mg/L nificantly associated with premenstrual mood symptoms, regardless of whether the conservative definition (symptom disappeared within 3 days of onset of menses) was used (OR=1.46, 95% CI 1.22–1.75) or if the symptom did not disappear within 3 days of onset of menses (OR=1.74, 95% CI 1.17–2.58) (Table 2). Similarly, in unadjusted analyses, hsCRP levels>3 mg/L were significantly associated with premenstrual abdominal cramps/pain, regardless of whether the conservative definition was used (OR=1.84, 95% CI 1.52–2.23) or if the symptom did not disappear within 3 days of onset of menses (OR=2.36, 95% CI 1.61–3.46). Also, in unadjusted analyses, hsCRP levels>3 mg/L were significantly positively associated with premenstrual appetite cravings/ weight gain/bloating, regardless of whether the conservative definition (OR=1.78, 95% CI 1.42–2.22) or less conservative definition (OR=2.30, 95% CI 1.54–3.42) was used. An elevated hsCRP level was not associated with re porting premenstrual breast pain or headaches in unadjusted analyses. Other factors related to each symptom group were 12 similar to those we found previously (data not shown). We also examined the unadjusted mean hsCRP by number of symptom groups reported and found a trend of increasing means (from 3.11–7.78 mg/L for none, 3.18–9.12 mg/L for one, 3.06–4.76 for two, 3.51–5.31 mg/L for three, 4.25– 6.52 mg/L for four to 4.22–for five symptoms)5.38 mg/L with increasing number of symptom groups, which was sig nificant in ANOVA (p=0.026), but the trend was not monotonic. However, because the distribution of hsCRP was skewed to the right, we examined median hsCRP by number of symptom groups reported and found that the median in creased monotonically from 1.0 mg/L for none to 2.1 mg/L for five symptoms reported. Further, the unadjusted ORs for the association of elevated hsCRP with number of symptoms
Independent variables and covariates a Age, (mean, SD) a hsCRP (mg/L) £3 >3 b Race/ethnicity African American Caucasian Chinese Hispanic Japanese c Education £High School >High School
<$35,000 $35–75,000 >$75,000 e Menopausal status Premenopause
2 f BMI (kg/m ) <18.5 18.5–24.9 25–29.9 >30 g Parity None 1–3 4+ h Smoke exposure Never smoker/no passive Never smoker/ some passive Former smoker/ no passive Former smoker/ any passive
n Antiinflammatory medications No 493 26.4 1377 Yes 225 21.0 844 o No. of comorbidities None 220 28.1 564 1 253 25.1 754 2 153 24.2 478
p Diabetes No 683 24.9 2057 Yes 26 17.6 122 q High blood pressure No 571 24.4 1765 Yes 136 24.6 417 r Osteoporosis No 703 24.7 2148 Yes 7 21.2 26 s Arthritis No 631 26.2 1779 Yes 78 16.2 403 t Fibroids No 578 25.1 1724 Yes 131 22.7 447 u Cancer No 700 24.7 2133 Yes 10 17.5 47
33.0 26.6
389 47
433 2
252 391
23.5 27.1
49.9 62.2
1208 292
50.1 1202 37.8 299
51.7 52.8
48.3 47.2
83.9 90.2
795 128
2021 434
1615 353
67.0 73.4
48.1 1470 51.9 45.6 31 54.4 (continued)
108 82 97 155
417 19
736 185
76.5 72.9
1967 487
84.2 88.1
609 146
25.3 30.4
74.7 69.6
428 8
Table1. (Continued) Appetite/weight/bloat Breast pain Headano. of symptomsches Total No Yes No Yes NoYes<3>3 noror % nor % ornor % ornoror % noror % noror % noror % noror % Mean SD Mean SD Mean SD Mean SD Mean SDMean SD Mean SD Mean SD 56 11.2 442 88.8 162 32.5 336 67.5 358 71.9140 28.1 221 44.4 277 55.6
67.3 69.8
1259 746
243 191 208 291
590 558 557 791
65.2 70.2 68.2 69.2
455 449 446 655
514 685 434 372
65.6 68.0 68.8 72.0
479 487 482 721
720 35
73.7 76.4
150 246 174 200
589 164
74.8 70.3
25.2 29.7
396 342 329 459
56.7 53.4 50.3 48.5
80.9 75.6 72.4 61.3
19.1 24.4 27.6 38.7
43.3 46.6 49.7 51.5
270 322 197 145
80.2 83.4 86.8 92.8
629 840 548 480
870 1250
81.1 86.5
368 449
203 195
55.8 45.0
474 795
44.2 55.0
noror % noror % noror % noror % SD Mean SD Mean SD